UNIVERSITY PARK, Pa. — The U.S. Department of Energy awarded a $29 million grant to seven multi-institution teams across the country to explore applications of machine learning, artificial intelligence and data resources in fusion and plasma sciences. A Penn State faculty member is one of the 19 individual recipients recognized, with a share close to $400,000 to focus on the use of machine learning to help mitigate nuclear reactor disruptions.

Romit Maulik, assistant professor in the Penn State College of Information Sciences and Technology (IST) will collaborate with researchers from Los Alamos National Laboratory, the University of Florida and The University of Texas at Austin (UT Austin) over the next three years with this funding. The team’s project is titled, “DeepFusion Accelerator for Fusion Energy Sciences in Disruption Mitigation.”  The researchers will focus on using machine learning to better predict and prevent imminent failures in nuclear fusion reactors, which generate energy through the same process that powers the sun.

“Artificial intelligence and scientific machine learning are transforming the way fusion and plasma research is conducted,” said Jean Paul Allain, associate director for fusion energy sciences within the DOE’s Office of Science, in a DOE press release. Allain is currently on leave from his role as the head of the Ken and Mary Alice Lindquist Department of Nuclear Engineering at Penn State. “The U.S. is leveraging every tool in its pursuit of an aggressive program that will bring fusion energy to the grid on the most rapid timescale.”

Before joining IST this year, Maulik — who is also a Penn State Institute for Computational and Data Sciences co-hire — had been collaborating on the Tokamak Disruption Mitigation project with Los Alamos National Laboratory to build machine learning algorithms to aid scientific discovery in nuclear fusion. He said this grant will support him as he takes a deeper dive into the machine learning side of things.

“Nuclear fusion reactors are prone to catastrophic performance failures,” Maulik said. “This creates a safety hazard that prevents nuclear fusion from being commercialized or becoming a power source for the grid.”

Maulik said one grand challenge is the inability to predict when a reactor will fail. Simulations provide insight, but they may be too slow and expensive to be used in real time to detect what might happen.

“We want to use data science to accelerate these simulations dramatically,” Maulik said. “If we can rapidly predict an imminent failure, we can control the factors that affect our experiment so that this failure may be avoided.”

Maulik said the project will develop machine learning models using previously run simulations as well as experimental data that is coming from actual reactor facilities.

“Once we’re able to detect failures ahead of time, we’ll be able to begin proposing mitigation strategies,” he said.

The work will benefit from the interdisciplinary and collaborative nature of the research, according to Maulik, who will provide machine learning expertise. Tan Bui-Thanh, associate professor in the Department of Aerospace Engineering and Engineering Mechanics at UT Austin, will contribute optimization and computational statistics knowledge. Chris McDevitt, associate professor of nuclear engineering at the University of Florida, and Xianzhu Tang, plasma physicist at Los Alamos National Laboratory, will be the physics experts on the project.

“We are all part of the DOE ecosystem and have spent years working together on projects and delivering results for Tokamak Disruption Mitigation research,” Maulik said. “The nature of this grant means that every cog needs to mesh with every other institution. We need to leverage our varied knowledge and come together to build something that is viable.”

Maulik said that complete development of a program capable of identifying and disrupting potential problems is likely to take more than three years to develop, but the team is hoping to make inroads. And they’re hoping to inspire future researchers to consider tackling the grand challenges of nuclear fusion.

“One of the things I hope to accomplish through this grant is to raise awareness among informatics and computer science students of the unconventional challenges that require machine learning, data science and information science skills,” Maulik said. “Nuclear fusion is perhaps something that’s not on their radar.”